_DSBEvaluations --> _SafetyEvaluation (Azure#39671)

* Adding 	asks arg to dsb evals

* adding unit tests and coherence evaluator

* adding evaluation_name param

* updates

* adding more unit tests

* updates

* make model config optional

* make evaluators empty by default

* add adversarial qa as default scenario

* add pyrit dependency and sample

* removing pyrit dependency for now

* updating conftest and test recordings

* ci please be happy with me

* update test recording

Author: slister1001

sdk/evaluation/azure-ai-evaluation/assets.json

@@ -2,5 +2,5 @@
   "AssetsRepo": "Azure/azure-sdk-assets",
   "AssetsRepoPrefixPath": "python",
   "TagPrefix": "python/evaluation/azure-ai-evaluation",
-  "Tag": "python/evaluation/azure-ai-evaluation_995c25afac"
+  "Tag": "python/evaluation/azure-ai-evaluation_398d07a53e"
 }

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_dsb_evaluation/__init__.py renamed to sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_safety_evaluation/__init__.py

@@ -0,0 +1,251 @@
+# coding: utf-8
+# type: ignore
+
+# -------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for
+# license information.
+# --------------------------------------------------------------------------
+"""
+DESCRIPTION:
+    These samples demonstrate usage of _SafetyEvaluation class with various _SafetyEvaluator instances.
+    
+USAGE:
+    python evaluation_samples_safety_evaluation.py
+
+    Set the environment variables with your own values before running the sample:
+    1) AZURE_OPENAI_ENDPOINT
+    2) AZURE_OPENAI_API_VERSION
+    3) AZURE_OPENAI_DEPLOYMENT
+    4) AZURE_SUBSCRIPTION_ID
+    5) AZURE_RESOURCE_GROUP_NAME
+    6) AZURE_PROJECT_NAME
+
+"""
+
+class EvaluationSafetyEvaluationSamples(object):
+    def evaluation_safety_evaluation_classes_methods(self):
+        import os
+        import asyncio
+        from azure.ai.evaluation._safety_evaluation._safety_evaluation import _SafetyEvaluation, _SafetyEvaluator
+        from azure.ai.evaluation.simulator import AdversarialScenario
+        from azure.identity import DefaultAzureCredential
+        # [START default_safety_evaluation]
+        def test_target(query: str) -> str:
+            return "some response"
+
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_default = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+        safety_evaluation_default_results = asyncio.run(safety_evaluation_default(
+            target=test_target,
+        ))
+        # [END default_safety_evaluation]
+
+        # [START content_safety_safety_evaluation]
+
+        def test_target(query: str) -> str:
+            return "some response"
+
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_content_safety_single_turn = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+
+        safety_evaluation_content_safety_single_turn_results = asyncio.run(safety_evaluation_content_safety_single_turn(
+            evaluators=[_SafetyEvaluator.CONTENT_SAFETY],
+            evaluation_name="some evaluation",
+            target=test_target,
+            num_turns=1,
+            num_rows=3,
+            output_path="evaluation_outputs_safety_single_turn.jsonl",
+        ))
+        # [END content_safety_safety_evaluation]
+
+        # [START content_safety_multi_turn_safety_evaluation]
+        def test_target(query: str) -> str:
+            return "some response"
+        
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_content_safety_multi_turn = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+
+        safety_evaluation_content_safety_multi_turn_results = asyncio.run(safety_evaluation_content_safety_multi_turn(
+            evaluators=[_SafetyEvaluator.CONTENT_SAFETY],
+            target=test_target,
+            num_turns=3,
+            num_rows=3,
+            output_path="evaluation_outputs_safety_multi_turn.jsonl",
+        ))
+
+        # [END content_safety_multi_turn_safety_evaluation]
+
+        # [START content_safety_scenario_safety_evaluation]
+
+        def test_target(query: str) -> str:
+            return "some response"
+
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+    
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_content_safety_scenario = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+
+        safety_evaluation_content_safety_scenario_results = asyncio.run(safety_evaluation_content_safety_scenario(
+            evaluators=[_SafetyEvaluator.CONTENT_SAFETY],
+            target=test_target,
+            scenario=AdversarialScenario.ADVERSARIAL_SUMMARIZATION,,
+            num_rows=3,
+            output_path="evaluation_outputs_safety_scenario.jsonl",
+        ))
+
+        # [END content_safety_scenario_safety_evaluation]
+
+        # [START protected_material_safety_evaluation]
+        def test_target(query: str) -> str:
+            return "some response"
+        
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_protected_material = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+
+        safety_evaluation_protected_material_results = asyncio.run(safety_evaluation_protected_material(
+            evaluators=[_SafetyEvaluator.PROTECTED_MATERIAL],
+            target=test_target,
+            num_turns=1,
+            num_rows=3,
+            output_path="evaluation_outputs_protected_material.jsonl",
+        ))
+
+        # [END protected_material_safety_evaluation]
+
+        # [START groundedness_safety_evaluation]
+        def test_target(query: str) -> str:
+            return "some response"
+        
+        grounding_data = "some grounding data"
+        
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_groundedness = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+        safety_evaluation_groundedness_results = asyncio.run(safety_evaluation_groundedness(
+            evaluators=[_SafetyEvaluator.GROUNDEDNESS],
+            target=test_target,
+            source_text=grounding_data,
+            num_rows=3,
+            output_path="evaluation_outputs_groundedness.jsonl",
+        ))
+
+        # [END groundedness_safety_evaluation]
+
+        # [START quality_safety_evaluation]
+        def test_target(query: str) -> str:
+            return "some response"
+        
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_quality = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+
+        safety_evaluation_quality_results = asyncio.run(safety_evaluation_quality(
+            evaluators=[_SafetyEvaluator.RELEVANCE, _SafetyEvaluator.COHERENCE, _SafetyEvaluator.FLUENCY],
+            target=test_target,
+            num_turns=1,
+            num_rows=3,
+            output_path="evaluation_outputs_quality.jsonl",
+        ))
+
+        # [END quality_safety_evaluation]
+
+        # [START xpia_safety_evaluation]
+
+        def test_target(query: str) -> str:
+            return "some response"
+        
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_xpia = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+
+        safety_evaluation_xpia_results = asyncio.run(safety_evaluation_xpia(
+            evaluators=[_SafetyEvaluator.INDIRECT_ATTACK],
+            target=test_target,
+            num_turns=1,
+            num_rows=3,
+            output_path="evaluation_outputs_xpia.jsonl",
+        ))
+        
+        # [END xpia_safety_evaluation]
+
+        # [START upia_safety_evaluation]
+        def test_target(query: str) -> str:
+            return "some response"
+        
+        azure_ai_project = {
+            "subscription_id": os.environ.get("AZURE_SUBSCRIPTION_ID"),
+            "resource_group_name": os.environ.get("AZURE_RESOURCE_GROUP_NAME"),
+            "project_name": os.environ.get("AZURE_PROJECT_NAME"),
+        }
+
+        credential = DefaultAzureCredential()
+
+        safety_evaluation_upia = _SafetyEvaluation(azure_ai_project=azure_ai_project, credential=credential)
+        safety_evaluation_upia_results = asyncio.run(safety_evaluation_upia(
+            evaluators=[_SafetyEvaluator.DIRECT_ATTACK],
+            target=test_target,
+            num_turns=1,
+            num_rows=3,
+            output_path="evaluation_outputs_upia.jsonl",
+        ))
+        # [END upia_safety_evaluation]
+
+if __name__ == "__main__":
+    print("Loading samples in evaluation_samples_safety_evaluation.py")
+    sample = EvaluationSafetyEvaluationSamples()
+    print("Samples loaded successfully!")
+    print("Running samples in evaluation_samples_safety_evaluation.py")
+    sample.evaluation_safety_evaluation_classes_methods()
+    print("Samples ran successfully!")

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_dsb_evaluation/_dsb_evaluation.py renamed to sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_safety_evaluation/_safety_evaluation.py

@@ -71,7 +71,7 @@
         "azure-identity>=1.16.0",
         "azure-core>=1.30.2",
         "nltk>=3.9.1",
-        "azure-storage-blob>=12.10.0",
+        "azure-storage-blob>=12.10.0"
     ],
     project_urls={
         "Bug Reports": "https://github.com/Azure/azure-sdk-for-python/issues",

sdk/evaluation/azure-ai-evaluation/samples/evaluation_samples_safety_evaluation.py

@@ -196,8 +196,9 @@ def evalutation_run_sanitizer() -> None:
         add_body_key_sanitizer(json_path="$..userTenantId", value=ZERO_GUID)
         add_body_key_sanitizer(json_path="$..upn", value="Sanitized")
 
-        # remove the stainless retry header since it is causing some unnecessary mismatches in recordings
+        # remove the stainless retry header and read timeout since it is causing some unnecessary mismatches in recordings
         add_batch_sanitizers({Sanitizer.REMOVE_HEADER: [{"headers": "x-stainless-retry-count"}]})
+        add_batch_sanitizers({Sanitizer.REMOVE_HEADER: [{"headers": "x-stainless-read-timeout"}]})
 
     azure_workspace_triad_sanitizer()
     azureopenai_connection_sanitizer()